Audit of hospitalization and ICU data

[jeanpaulrsoucy]

Instead of using covid19tracker.ca data, we should use the hospitalization/ICU data we have collected in-house (e.g., SK) and simply discontinue reporting when public reporting was discontinued. We could also look at the CCCS dataset to see if anything useful can be obtained (#67).

• [x] Report PHAC country-level hosp and ICU data as a separate dataset
• [x] AB hosp and ICU (24a572ea-0de3-4f83-b9b7-8764ea203eb6) (two datasets: figure 9 and figure 10)
• [x] BC hosp and ICU (91367e1d-8b79-422c-b314-9b3441ba4f42)
• [x] PE (from news releases and webpage)

The datasets still relying on covid19tracker for early data are:

• [x] BC (2020-01-25–2021-03-12) (partial replacement)
• [x] MB (2020-01-25–2021-02-03) (replaced)
• [x] NB (2020-01-25–2021-03-07) (partial replacement)
• [x] NL (2020-03-17–2021-03-15, ICU data only) (no change)
• [x] NS (2020-01-25–2021-01-18) (replaced)

[jeanpaulrsoucy]

There are three issues here:

• Not all provinces report current hospitalizations/ICU.
• Some provinces distinguish between hospitalizations "with" or "for" COVID. For example, Alberta has a "with" time series covering the whole time series, as well as a "for" time series beginning in 2022. This would be a great use case for "grouping" variables to provide more data.
• Of course, the definition of "with" not "for" is that it lacks the typical COVID presentation, which doesn't technically exclude COVID as a cause. Additionally, some hospitalizations related to COVID are not captured because they are due to the downstream sequelae of COVID.
• Some provinces report new hospital/ICU admissions. This could be a separate dataset, for those that report it.

[jeanpaulrsoucy]

For Manitoba, our current time series harvested from 8cb83971-19f0-4dfc-b832-69efc1036ddd uses the Active_Hospitalizations and Active_ICU_Patients fields. However, it should probably use the Total_Hospitalizations and Total_ICU_Patients fields, as it appears the "active" refers to those with an active infection. However, patients in "total" but not "active" are still there due to COVID-19 at admission, so should be included. It also appears this is the value used by covid19tracker.ca.

See language discussing the differences in this article: https://winnipeg.ctvnews.ca/covid-19-hospitalizations-in-manitoba-down-icu-cases-and-deaths-increase-on-friday-1.5825152 this bulletin: https://news.gov.mb.ca/news/index.html?item=50517&posted=2021-01-31

there are 108 people in hospital with active COVID-19 as well as 151 people in hospital with COVID-19 who are no longer infectious but continue to require care, for a total of 259 hospitalizations

[jeanpaulrsoucy]

Early Manitoba COVID-19 hospitalization/ICU data could probably be reconstructed from the COVID-19 Bulletins (e.g., https://news.gov.mb.ca/news/index.html?item=48367&posted=2020-05-31).

[jeanpaulrsoucy]

The first step to the audit should be removing covid19tracker.ca data that cannot be sourced (e.g., territories, MB) or may represent a different quantity (e.g., admissions rather than occupancy, which seems to be what happened for NL), or is otherwise no longer updated (i.e., reporting stale data).

[jeanpaulrsoucy]

From the BCCDC dashboard dataset (91367e1d-8b79-422c-b314-9b3441ba4f42), there are the following data notes:

As of January 14, 2022, the data source for “Currently Hospitalized” has changed to better reflect hospital capacity. Comparisons to numbers before this date should not be made. Hospital data include admissions for people who test positive for COVID-19 through hospital screening practices, regardless of the reason for admission. Therefore, reprted hospitalizations overestimate the true number of people who are hospitalized specifically due to COVID-19 infection. "Currently Hospitalized" is the number of people who test positive for COVID-19 through hospital screening practices, regardless of the reason for admission, as recorded in PCMS on the day the dashboard is refreshed. It is reported by the hospital in which the patient is hospitalized, rather than the patient's health authority of residence. Critical care values (intensive care units, high acuity units, and other critical care surge beds) include individuals who test positive for COVID-19 and are in critical care, as recorded in PCMS

This data notes page on the old dashboard also contains other relevant notes to capture about case and death data (#37).

[jeanpaulrsoucy]

Important context for PEI's hospitalization data (from their 2022 annual summary report):

[jeanpaulrsoucy]

For pre late-2021 hospitalization data for PEI, we need to look through news releases:

• https://www.princeedwardisland.ca/en/search/site/COVID-19%2520hospital?f%5B0%5D=type%3Anews&f%5B1%5D=field_news_type%3A22
• https://www.princeedwardisland.ca/en/news/prince-edward-island-reports-first-covid-19-hospitalization

We may only have enough to report total admissions rather than occupancy until late 2021.

[jeanpaulrsoucy]

Possibly make a note about for/with distinction for hospitalizations being available for NB but only at the very end of the time series we use (40d93d2b-5c85-4592-b844-fb78b5ae5cec). Probably not worth incorporating.

It is also not clear if NB hospital numbers for this dataset include those in ICU. The first data point on 2020-09-20 with hosp: 23 and ICU: 14 would be quite gnarly if ICU cases make up over half of total hospitalized cases.

Notably: once they switched to reporting hospital occupancy in weekly reports, the report did explicitly say the hospitalization number did include those in ICU.

This news story gives 20 in hospital, with no mention of ICU. Comparing to the dashboard dataset of the time (4f194e3b-39fd-4fe0-b420-8cefa9001f7e), the dataset gives 20 in hospital, 13 in ICU on 2021-04-08. This would give support to the hospital number being inclusive of those in ICU.

[jeanpaulrsoucy]

Note regarding ON ICU occupancy data:

Due to a system outage, ICU data for September 9-13, 2023 and September 21, 2023 are not available

This seems to also be the case for 2021-07-17.

Need to also verify whether icu_crci_total may be used for this value rather than adding up icu_current_covid and icu_former_covid.

Update (2023-09-29): ON has retracted all of its updated hosp and ICU data:

Due to technical issues, data for September 9, 2023 onwards are not currently available.

[jeanpaulrsoucy]

AB hospital data have been updated, using the "for COVID-19" dataset after 2022-02-01. This should be noted in the data notes.

[jeanpaulrsoucy]

News releases may be available for older hospitalization data:

• BC: https://news.gov.bc.ca/Search?Ministry=Health&q=COVID-19%20pandemic%20update&Page=28
• NB: https://www2.gnb.ca/content/gnb/en/news/search_results.html?_charset_=UTF-8&newsFromDate=DD%2FMM%2FYYYY&newsToDate=DD%2FMM%2FYYYY&newskeyword=COVID-19&newsdeptId=&start=900&hits=25
• NS: https://novascotia.ca/news/search/?deptnum=0&month=&year=2020&q=COVID-19&condition=AND&page=43

[jeanpaulrsoucy]

Status of replacement of covid19tracker datasets:

• BC: Partial replacement of covid19tracker dataset for hosp occupancy before 2020-03-17 (first date where covid19tracker dataset has unique data; news releases are less frequent than the unarchived dashboard dataset covid19tracker was pulling after this date as well)
• NB: Partial replacement of covid19tracker dataset before 2020-06-30, when NB transitioned to more frequent updates using its unarchived dashboard (which covid19tracker was pulling from)
• NS: Complete replacement of covid19tracker dataset using daily news releases

[jeanpaulrsoucy]

Updating some historical data:

• Censor BC hosp/ICU occupancy data on the following dates: 2021-04-01, 2021-05-05, 2022-04-28, 2022-08-18
• Censor ON ICU occupancy data from the following date: 2020-07-03

[jeanpaulrsoucy]

Final tasks - investigating CCCS data (#67):

• [x] Confirm CCCS data do not contain useful info for NT
• [x] Confirm CCCS data do not contain useful info for NU
• [x] Confirm CCCS data do not contain useful info for YT

And confirming whether the SK time series should be partially replaced with the report data that distinguish between "for" and "with".

[jeanpaulrsoucy]

Report on CCCS data (#67), which includes info on admissions (#111):

NT:

date	region	hosp	icu
2021-08-18	NT	1	0
2021-08-20	NT	1	0
2021-08-23	NT	1	0
2021-08-25	NT	12	4
2021-08-27	NT	12	4
2021-08-30	NT	13	4
2021-09-03	NT	14	4
2021-09-10	NT	23	6

This is simply the cumulative admissions data that gets reported on the dashboard. The main difference is that the data on the dashboard are behind:

NT | 2021-09-09 | 17 | 6

The reason for this is no reporting before ICU went from <5 -> 6 and that they have added ICU admissions to hospital admissions (making the assumption that hospital admissions correspond to non-ICU admissions only). It's not clear if this is accurate or not: the NT dashboard offers no help, nor does any of the reporting of the data that I can find. CIHI data indeed suggest higher hospital and ICU admissions by March 2022, but we also know there are differences in how CIHI and the provinces count admissions (e.g., see this article). I have confirmed that the ICU admissions numbers sometimes increase without a corresponding increase in hospital admissions, but this could be because someone in transferred from non-ICU to ICU.

The CCCS time series resets on 2022-01-13, with notes referencing reporting on active hospitalizations. However, later reports revert to counting the number of admissions. The time series resets again on 2022-03-28, and the same pattern repeats.

Conclusion: leave existing NT admissions time series alone but with a data note. Do not add hospital/ICU occupancy data.

NU:

NU has a very brief run of hospital admission data based on news releases between 2022-01-26 and 2022-03-09. These data are too fragmented and inconsistent to be totally clear from which date they represent cumulative hospitalizations; one article linked also mentions data revisions:

There have been nine hospitalizations in Nunavut due to Covid-19, some past cases reported as Covid-19 hospitalizations were previously corrected and taken off the list.

Later articles mention specifically the Omicron wave, but again the brief duration and lack of clarity make it not worth including.

On Tuesday, the territory’s total hospitalizations due to this wave of the COVID-19 pandemic rose from 23 to 27, though little information has been shared on the specifics of each hospitalized case, including how many discharges there have been.

Conclusion: Do not include admissions or occupancy data for NU.

YT:

The time series begins with cumulative hospital admissions but resets on 2021-11-08 and continues until 2022-08-25. This series of values due appear to be active hospitalizations, but they lack data notes until later. Values are pulled from news releases until the 2022-02-18 data note mentions the dashboard now reports active hospitalizations:

YT's official website is now informing the number of hospitalizations, which is 0 as of February 18th.

It links to this dashboard page. Of course, this page is now defunct and links to the archived datasets. However, near as I can tell, this page only ever reported daily admissions, which is supported by the Wayback Machine's version of the "Definitions" webpage.

Conclusion: Do not include admissions or occupancy data for YT.

[jeanpaulrsoucy]

For the weekly SK report dataset on hosp/ICU occupancy, the hospital and ICU overall occupancy values are reported, but the hospital values are also broken down as: Total COVID Related Illness, Total Incidental COVID Infection, Total Undetermined. However, ICU data are not broken down in a similar way.

The source dataset decomposed hospital occupancy into 'Total Covid Related Illness', 'Total Incidental Covid Infection', and 'Total Patient Under Investigation'. However, ICU occupancy was not decomposed in this way. Thus, we report 'Total Covid Hospitalized' and 'Total Covid Adult ICU/ICU Surge' for consistency and to allow fair calculation of non-ICU hospital occupancy.

These columns can be included as an extra dataset (#125), and a note can be added, but these data will not be used as a partial replacement.

[jeanpaulrsoucy]

I can't find the original comment about the MB bulletin data, so just for the sake of preservation:

We can reconstruct historical MB hospital and ICU occupancy data, which seems to have been reported in their COVID bulletins between:

• Bulletin 36 on 2020-03-30 (https://news.gov.mb.ca/news/index.html?item=47338&posted=2020-03-30)
• Bulletin 551 on 2022-01-07 (https://news.gov.mb.ca/news/index.html?item=53158&posted=2022-01-07)

The terminology for reporting seems to have changed at least once. In the first example, we see:

• four individuals are currently hospitalized,
• one individual is currently in intensive care,

In the final example, we see:

• 297 Manitobans hospitalized with COVID-19 including 257 people with active COVID-19 as well as 40 people who are no longer infectious;
• a total of 34 Manitoba patients receiving intensive care for COVID-19 including 33 people in intensive care units with active COVID-19 as well as one person who is no longer infectious but continue to require critical care;

In the latter case, we are interested in the total hospital and ICU occupancy numbers (i.e., 297 and 34, respectively).

[jeanpaulrsoucy]

Should check news releases for ICU data for NL:

https://www.gov.nl.ca/releases/covid-19-news/

[jeanpaulrsoucy]

Decided against using data from the NL news releases (https://www.gov.nl.ca/releases/covid-19-news/) for ICU occupancy data. These news releases start later and report less frequently than the NL news releases (in early April, they not doubt start drawing from the dashboard), so there is no real advantage to trying to build this dataset.